AG-CLIP: Attribute-guided CLIP for Zero-shot Fine-grained Recognition

Key Points

• Explore the use of Attribute-guided CLIP for enhancing zero-shot fine-grained recognition capabilities.
• Developed a novel clustering algorithm integrating attributes into the CLIP framework
• Applied the algorithm to various datasets for evaluation
• Performed comparative analysis with baseline models
• Demonstrated significant improvement in fine-grained recognition accuracy
• Achieved effective zero-shot learning performance
• Highlighted versatility across multiple datasets